The investigation of a crime involving a gun often includes an analysis of the grooves carved into the bullet as it travels the length of the gun's barrel. Typically, experts visually compare the grooves between two bullets or a bullet and a gun barrel. When the database is large, however, the time to reach a definitive result can impede the investigation. Researchers at the Universität Karlsruhe (Karlsruhe, Germany) and the Technische Universität München (Munich, Germany) have developed an image-processing strategy aimed at automating the analysis and comparison of bullet grooves.1 This approach will also aid in identifying and matching tool marks left when a thief breaks into a car or building.
Rather than trying to compare two-dimensional (2-D) images of grooves as a human might do, a one-dimensional (1-D) trace of a groove pattern serves best for machine identification—not only is a 1-D trace simpler, but it can be an average of many 1-D scans across the grooves, thus decreasing random effects. But because grooves left from rifle barrels or burglary tools are not often perfectly straight, obtaining a compiled 1-D
scan is not straightforward—the object creating a groove may nonlinearly translate, rotate, or both as it moves relative to the surface being scored. In addition, capturing an evenly illuminated 2-D image of a nonplanar surface (to be collapsed into a 1-D scan) is not trivial.
Given a certain surface shape (for example: cylindrical surface and straight grooves, planar surface and curved grooves, or curved surface and curved grooves), the German researchers first acquire a 2-D image-striation pattern using an image-fusion approach. In the instance of a curved metal surface—which, because it combines specular reflection and scatter, is difficult to illuminate—a series of images illuminated from different angles is taken. To create a fused image of the surface, the portion of each image that shows highest contrast (based on automatic analysis) for that angle of illumination is excised—these sections are combined in a final, uniformly high-contrast image. Further preprocessing with a Gaussian high-pass filter eliminates slowly varying gray-scale fluctuations.
Snakes and signatures
Next, the 1-D compilation, or "signature," is obtained. To do this, a groove-straightening algorithm—guided initially by a human examiner, who provides rough orientation by marking several points on a prominent groove—determines an accurate groove contour, or "snake," by finding groove edges. Adjusting the 2-D groove pattern by sliding each 1-D cross-sectional slice by the groove deviation at that point produces a 2-D pattern of straight grooves that can be collapsed into a 1-D signature (see figure). Rotations of the scoring object as it travels were found by the researchers to be insignificant in most cases.
The experiments were carried out on striated surfaces produced by a German forensic institute, according to Michael Heizmann of the Universität Karlsruhe. The surfaces have nothing to do with rifles or bullets, he notes. Instead, they contain marks generated from screwdrivers, as the institute wanted to make available a test collection of striation marks that show generalized properties and difficulties.
The algorithm has proven to be insensitive to dust particles and considerable distortions to the 2-D images, although discontinuities in the grooves can cause problems. Once the image is collapsed, the signature can be analyzed and compared to other signatures using correlation methods.
REFERENCE
- M.Heizmann and F. Puente León, Opt. Eng. 42(12) (December 2003).